Latest / CareTalk: Healthcare. Unfiltered. / How AI Is Changing The Way Doctors Think w/ Dr. Bapu Jena, Harvard Medical School

Transcript

How AI Is Changing The Way Doctors Think w/ Dr. Bapu Jena, Harvard Medical School · CareTalk: Healthcare. Unfiltered.

• 0:07
  Welcome to Care Talk, America's home for incisive debate on healthcare, business, and politics.
• 0:13
  I'm John Driscoll, the chair of the UConn Health System, and a, a, a sometimes partner with, uh, our guest today in a, in a firm called JSL, which is an early-stage healthcare venture capital firm.
• 0:25
  And Bapu is a return, uh, visitor to the show.
• 0:29
  Bapu, maybe you could give people a little bit of a sense of your qualifications and your orientation as a, a patient, an expert, and, and sometimes a productive critic of healthcare based at Harvard.
• 0:42
  I'm trying to, I'm trying to use less of it and, and supply more of it.
• 0:45
  Uh, but, you know, thank you for having me, John.
• 0:47
  So my background is I'm an economist and, uh, a medical doctor.
• 0:50
  I, I did my PhD in economics, and I work mostly in, on topics what we call health economics.
• 0:57
  And within that field, I have a particular interest in the freakonomics of medicine, which is thinking about clever and quirky ideas in medicine as they apply to patients, as they apply to doctors, maybe as they apply to organizations within healthcare.
• 1:11
  And then I see patients at Massachusetts General Hospital in, in Boston, uh, several weeks out of the year.
• 1:17
  So- I didn't realize you were still practicing.
• 1:19
  Well, yeah.
• 1:20
  Most people don't.
• 1:20
  Yeah.
• 1:21
  I, I was, I was just, just on service, uh, uh, four weeks ago.
• 1:25
  So I, I, I… I'm always on service in the fall because I teach an undergraduate, uh, course at Harvard College on healthcare, and I like to do it in the fall because that allows the students to come and visit the hospital, uh, in the morning.
• 1:37
  So I still do it.
• 1:38
  So are you more of a real doctor and a fake economist or a real economist and a part-time doctor?
• 1:44
  Yeah, a little bit of both.
• 1:47
  Well, so maybe we could talk a little bit about, you know, uh, you-- A lot of folks have been really interested in the way AI is taking over the world.
• 1:59
  Uh, and there's been a lot of pushback on AI in a lot of parts of society, but, but surprisingly to me, a real embrace of artificial intelligence, particularly tools like OpenEvidence and Doximity, by doctors and clinicians and, and, and honestly students as well, folks who are, uh, training to be doctors, but even, even, even the experts.
• 2:21
  Maybe you could tell-- talk a little bit about, uh, how quickly, um, docs have embraced AI and kind of what the implications of those are.
• 2:30
  Uh, uh, so I'd say two things.
• 2:31
  One is very quickly.
• 2:33
  Uh, if I think about healthcare technologies and the pace of adoption You know, outside of COVID-19 vaccines, which were incredibly quickly adopted, uh, the AI tools that physicians are using, including something like Open, Open Evidence, um, is enormous.
• 2:48
  We, we just had a paper in, uh, JAMA come out last week, and if you just look at physicians, let's say internists, by our estimate, something like 80% or more of internists have used Open Evidence at some point.
• 3:02
  And that's not to mention other tools, um, that, that clinicians use for, for support, clinical decision support.
• 3:09
  So the uptake is enormous, and I, I think the reason why is if you contrast, let's say, electronic health records, where the uptake eventually became large but was slow and painful, is that these tools, these AI tools, clinicians clearly think that they're offering clinical value to them in a way that they didn't necessarily think that, uh, electronic health records were.
• 3:30
  They thought those were more of sort of a hassle than serving a core clinical function for them.
• 3:36
  The second thing is these tools are largely free, whereas electronic health records were, were very expensive.
• 3:41
  And then I think the last thing is, if you've used an electronic health record, you know that they're not easy to use.
• 3:46
  The user interfaces- They're the opposite- Opposite of that … of easy to use.
• 3:50
  Whereas these, these tools, these AI, LLM, LLM, uh, large language model-based, uh, clinical decision support tools are incredibly easy to use.
• 3:59
  And so it's the, the pace I think is driven by the ease of use, the value that it provides clinically, and, and the cost.
• 4:07
  It's free.
• 4:08
  But I think, I think what's interesting, Bapu, is that the 80% of docs admit to using it.
• 4:15
  I think the rest are lying.
• 4:16
  I mean- Yeah.
• 4:17
  … e-everybody, um, uh, it, it's remarkable how common and extensible these, you know, access is and, and how it-- without really any filters or authorities, um, doctors are jumping in.
• 4:34
  Is there-- Are, are you, are you okay with that as a clinician?
• 4:39
  Uh, in so many other areas of healthcare, um, technology has moved at the speed of trust and cost versus, uh, power s- you know, sort of value.
• 4:53
  Uh, and in this case, it, it seems to have sort of, uh, just become embedded overnight.
• 4:59
  Is that something we should be afraid-- Is, is AI something we should be worried about?
• 5:04
  Or should we sort of embrace what, what feels like the inevitable of AI being thought everywhere and embraced completely in healthcare?
• 5:11
  I think in healthcare, at least in the area where I'm most familiar, which is using it to, to help clinicians make diagnoses, refine treatments that they offer patients, I, I think that the, the value is really, um, it's really in-incredible.
• 5:28
  And there's a difference here, which is that when a physician is using a tool like this, they already have a lot of expertise.
• 5:36
  So it's not as if you have someone who has absolutely no knowledge about medicine or clinical care asking, uh, a tool a question and then following it, you know, blindly.
• 5:45
  That's not what's happening here.
• 5:46
  You have people who already have a lot of expertise who are using it to refine, uh, their skills and decision-making.
• 5:53
  And so that's an important backstop that I think some people sometimes forget.
• 5:57
  Like this is, these are people who know what they're doing who are using it.
• 6:00
  The second thing is that there are, are sort of what I'd call kind of market mechanisms at play here, both at the individual physician level.
• 6:07
  Like if it's not offering you value, if it's giving you wrong decisions, you're gonna stop using it.
• 6:13
  And the fact that we see more and more people using over time, it's not like a drug, right?
• 6:16
  It's not addictive.
• 6:17
  It's something that you use more and more of because you realize it's offering some, some, some benefit, um, to you.
• 6:24
  Um, and then the last thing is, uh, unlike other sorts of things, there's a lot of interest in benchmarking different tools against each other and against expert clinicians.
• 6:34
  And, and all of these sort of exercises are demonstrating that there is a lot of benefit to be had.
• 6:40
  So I think there's a lot of convergence on, on positive things.
• 6:44
  Well, I think, uh, what I, what I worry about, Papu, is, and, and again, I'm just the, the, the podcast guy who's on a lot of boards and, and advises healthcare institutions like UConn and Johns Hopkins, is that these are, um, you know- Incredible data crunching machines, but they are a little bit, um, a function of the, all of the, of what's available on the internet.
• 7:13
  And I know of one, uh, internist who was looking for, um, where the CDC was on vaccinations and going to sort of the, the, the, the COVID vaccine, the flu vaccine, some of the other, um, vaccines.
• 7:31
  And, and the initial read from OpenEvidence was not accurate or complete.
• 7:37
  Uh, he actually had to go to-- Now, the great thing about OpenEvidence is you can actually go to the sources, and then he realized that there were some sources that were omitted, and he sort of got to the right answer, but only by applying his own analytics.
• 7:51
  The flip side is a, a friend of mine who's an MD-PhD, or was an MD-PhD with a very rare and exotic form of cancer at UCSF, was using the hallucinations of the models, of the L- the L- the frontier models, not the OpenEvidence, to actually come up with novel approaches to his own care- Mm-hmm … that Sloan Kettering, you know, one of the great institutions, cancer institutions, hadn't come up with.
• 8:15
  And so it's, uh, the two extremes that you could actually generate through what other people see as mistakes, potential novel approaches, and yet some of the basics are incomplete.
• 8:26
  How do you sort through that as a clinician?
• 8:30
  So I think that the second thing to me is actually, it's the most interesting, though maybe not the most important, and I'd say the following.
• 8:38
  So these tools have at least two big possibilities.
• 8:41
  One is to allow us to do our best with the knowledge that currently exists in the world.
• 8:48
  So, and it's impossible for a physician to know everything about every single disease, every single treatment, whereas a computer might be able to, to do that.
• 8:56
  And so in the first instance, we're talking about the possibility that a computer could, could aid clinicians in making diagnoses, and sometimes they'll make errors, but sometimes human beings will make errors as well.
• 9:07
  Um, and e- and everybody's thinking about that.
• 9:10
  But I think what people are thinking less about is that second angle, which is what you described, which is can, can these tools also help us generate new knowledge?
• 9:18
  And I think that they can in two places.
• 9:20
  Um, so your, your friend's experience is sim- something similar to my own.
• 9:23
  About, uh, about a year and a half ago, I remember I was on service, uh, at MGH, and we had a patient who had, um, uh, bad liver disease and who was nearing death.
• 9:34
  And their, their actual, their only wish was to, to be able to get home, to die in your home, but they were, they were so critically ill, it wasn't even clear if we could get him to do that.
• 9:43
  Um, and w- they had tried a number of different medications that were evidence-based, you know, based on clinical trials and clinical guidelines to try to address a particular clinical problem, and it wasn't enough.
• 9:58
  It wasn't good enough to get this person back home.
• 10:01
  And I remember sitting down with the team and saying to them, "Look, we should just open up ChatGPT," which is what people were using at the, at the time for this sort of stuff.
• 10:09
  Said, "We should say, 'Look, here's a clinical situation.
• 10:11
  What ideas do you have based on things that could physiologically just make sense?
• 10:16
  Maybe they haven't been studied, probably they haven't been studied, but what would you come up with?'" And the point here is, you know, the body is complex, but there's a lot of data and science out there that might help you come up with ideas that haven't been tested or proven before.
• 10:31
  And it came up with, like, really good ideas, and we could then go to the family and say, "Look, here are some things we could try. They're not based on any evidence right now, but in your situation, given, given, you know, what your preferences are, the goal is just to make it out of here," their willingness to try new things might be different.
• 10:49
  And I, you know, I've kinda come to appreciate the ability of these tools to offer new insights as opposed to reinforcing insights that you might not know but that a, a really expert clinician might know, which is a different function altogether.
• 11:04
  So I, I think there's a lot of power to generate new ideas here.
• 11:08
  So maybe there's two buckets, Papu.
• 11:10
  I wanna make sure I'm hearing you correctly.
• 11:12
  One is that The reality is that with medical research and n- innovation, we're generating every few years as much information about healthcare and, and research and innovation, um, as, as previously would've been, you know, every few years what would've been consumed decades.
• 11:35
  And that's just a, a, a massive amount of information that leveraging AI, doctors can, can actually access and get into workflow.
• 11:44
  But the other piece that you're suggesting is Actually leveraging the tool's own sort of, uh, own capability- Mm-hmm … to generate novel solutions.
• 11:59
  And I hadn't, I hadn't thought about that as sort of a nor- normal course thing, and you're saying we should l- you're kind of recommending that docs not just use this to consume actionable information based on what we know, but really actionable approaches that maybe we haven't tried.
• 12:18
  Exactly.
• 12:18
  It's, it's like brainstorming.
• 12:20
  I, I think, you know, part of, John, like what I'm probably best known for is coming up with these weird ideas and, and people will say, "Well, where did you come up with that?" And it's basically, it's basically linking two things, connecting two things that someone hadn't linked before.
• 12:33
  And may- maybe the best example I'll give is this, this paper we had years ago about marathons and mortality.
• 12:39
  People are more likely to die in a city that hosts a marathon because they can't get to the hospital because the roads are closed.
• 12:45
  Very obvious and intuitive idea when I tell it to you, but nobody had thought that marathons could, could cause increases in mortality.
• 12:53
  But a computer could do that because a computer would say, "All right.
• 12:57
  Well, we, we know that delays in care might affect outcomes. What are the things that delay care?" Very easy.
• 13:04
  Pope coming to visit, president coming to visit, marathons being held.
• 13:09
  Like, the computer would come up with those things in a way that a human being wouldn't necessarily draw those connections very easily.
• 13:15
  And so I think what, what… And, and, you know, the instance that, that I had used it for this specific purpose, I was basically asking it to draw connections between things based on some normal established physiologic or- How did you c- how did you come up with that?
• 13:31
  I mean, it, it is obvious.
• 13:32
  Yeah.
• 13:33
  Yeah.
• 13:33
  Well, I- So congratulations on g- congratulations on getting tenure for o- an obvious point.
• 13:38
  Exactly.
• 13:38
  I mean, not everybody could have done that, Bapu.
• 13:40
  Exactly.
• 13:40
  But how did you… But it is, it is really creative.
• 13:43
  How, where, how, what was the, how'd you, how'd you start there?
• 13:45
  What, what's the beginning?
• 13:46
  It was, it was my wife.
• 13:47
  So my wife was running a race.
• 13:48
  It was like a half marathon, and she wanted me to watch her on the race route, and it happened to go by Mass General Hospital.
• 13:55
  And so I drove to MGH planning to park at the hospital, uh, and watch her on the race route.
• 14:02
  But as I was driving there, the road was blocked.
• 14:05
  And so there, the main entrance to the hospital that, the off of Storrow Drive, which is our main thoroughfare, it was blocked, so I turned around and went back home.
• 14:13
  And I s- I saw her a few hours later and I said to her, I told her what happened, and she made this offhand comment.
• 14:19
  She said, "Well, what happened to everybody that needed to get to the hospital that day?" And that's what the idea was.
• 14:24
  It was kind of, it was in, in a big sense her idea, but most people wouldn't think of that comment and think, "Oh, there's an interesting research paper there." So, you know, I, I, that's what I for some reason do.
• 14:36
  I, I think it's probably because I have a lot of practice doing it, but I will sort of connect things.
• 14:40
  I think you could get some credit for being curious- Yeah, curious… and, and, and, and inventive.
• 14:45
  It's, it's-- I like- So, so you just went through kind of a, a I would think is traumatic experience with your mom's illness and passing, so I'm sorry for that.
• 14:54
  But that brings you right into it in the healthcare system as a sort of a, a creative critic and analyst as well as a practitioner, as well as, you know, part of the patient, you know, family.
• 15:09
  Any insights into how people are leveraging AI right now, and what some of the gaps in critical care for, for, for loved ones are right now that we still need to think about solving either through more humans or better technology?
• 15:27
  Mm-hmm.
• 15:28
  So I would say J- you know, I've spent my entire life being on the, on the other end, which is walking through a hospital hallway, going into someone's room, interacting, uh, with them for, you know, several minutes, and then walking out of that room and, and having to You know, having to be able to separate the severity and the seriousness of what they're going through from anything else that you might be thinking about in your life.
• 15:55
  Because if, if that, if you allow that to weigh on your mind every time you walk into a room, you wouldn't, you wouldn't be able to function.
• 16:01
  Um, but then the reverse is true when you yourself are either a patient or your family member.
• 16:07
  In this case, my mom was a patient.
• 16:09
  And I would say that the scarcest resource… You know, we think about resources in, in healthcare, and people focus on how much, what percentage of our GDP we spend on healthcare, how much drug costs, how much hospitals cost, all this stuff.
• 16:24
  Economists, policymakers focus on all that stuff.
• 16:27
  And I have sort of over the last couple years come to the realization, at least for me, that the scarcest resource is actually time.
• 16:36
  You know, a physician can spend 15 minutes with a patient.
• 16:40
  That's n- that's nowhere near enough.
• 16:42
  Like if you have a, if you have a loved one with dementia and you go into your PCP's office with your mom or your dad, and all they have is 15 minutes and that's not enough to provide comfort to somebody.
• 16:53
  It's enough to solve maybe a medical problem.
• 16:56
  May not even be enough for that.
• 16:58
  But I think that, you know, we have become very constrained in our ability to spend that resource, which is, which is time.
• 17:07
  And you know, nowadays when I'm in the hospital, I kinda think about the way I interact with people differently.
• 17:13
  Like I used to go into hospital rooms and kind of focus mostly on the medical stuff, and that's obviously important, but there's part, large parts of the team who do that.
• 17:22
  I actually just try to go in there and, and think to myself, "My goal is that this person should feel better when I leave this room than when I enter the room." And that, that's my goal.
• 17:34
  But you're on, you're on something that I think is bigger than time and more central to humans, which is, is kind of bringing your heart to the center of care- Yeah not just your head.
• 17:43
  And I think that's something I think that because of the way we've organized medicine is some, in the US, I think is sometimes Better handled in more traditional societies and actually less complex systems than we necessarily do in our, with our kind of medical factories, which are incredibly great at, in, in technology and teams and interventional stuff, but don't always get at the heart part of, uh And, and that's, that's part of healing.
• 18:12
  Yeah.
• 18:12
  And it's, and it's honestly like when you talk about burnout, it, it-- What physicians want to do is spend time at the bedside and feel like they are helping someone be better.
• 18:23
  I, I talk to the residents and medical students and undergrads who are thinking about going into medicine, and some of them think about medicine, some of them think about finance or consulting, and I say, "You know, all these jobs are, are great in different ways, but the one thing that, you know, in, in medicine," or let's say, "In finance, you, you cannot get the feeling of being able to go into someone's room and change their life." That you can't get that feeling in, in, uh, maybe in some ways you can, but that is unique- You can't.
• 18:54
  You can.
• 18:54
  You can.
• 18:54
  Okay, let's say you can.
• 18:55
  Let's just le- we'll leave it at that.
• 18:56
  I'll leave it at can.
• 18:57
  But it's unique to medicine, and that part is the part that's becoming increasingly de-emphasized.
• 19:02
  Like, that's, that's the fuel for a good doctor, is to be able to, you know, in- take that.
• 19:08
  That's the form of compensation that matters most to us, and it's being taken away slowly.
• 19:15
  Uh, and, and so, you know, you asked about AI, like I actually think that the link here between things that increase our productivity, like AI, is to, to, to make aspects of the job that are, are sort of less desirable or time-consuming, whether it be billing stuff, note-taking, um, searching for information around a diagnosis that you could get more rapidly.
• 19:38
  If those things allow physicians to spend more time ultimately with patients, then whether or not they change the quality of diagnoses, it's fine.
• 19:47
  If it allows you to spend more time with patients, then it's a win.
• 19:51
  More time with patients.
• 19:52
  I love it.
• 19:53
  That's-- I think we'll leave it there.
• 19:55
  Um, uh, I'm John Driscoll, the chair of UConn Health, co-host of "Care Talk." If you like what you heard or you didn't, we'd love you to subscribe on your favorite service.
• 20:07
  Uh, Dr. Jha, he- doc, healthcare economist, and perha- hopefully a futurist about the heart-centered aspects of health, brought to you through AI, a remarkable way to end it.
• 20:21
  Thank you, Bapu.
• 20:22
  Thank you, thank you, thank you.